Non-intrusive fault detection in reciprocating compressors

Abstract

This thesis presents a set of techniques for non-intrusive sensing and fault detection in reciprocating compressors driven by induction motors. The procedures developed here are "non-intrusive" because they rely only on electrical measurements to reconstruct the mechanical signals internal to the compressor. This allows for easy and non-intrusive determination of many fault sensitive signals that usually require complicated, expensive, and time consuming operations to measure. A sample of the signals produced by the procedures of this thesis are estimates of the cylinder suction and discharge pressures and a composite torque signal containing the effects of the mechanical loads within the compressor. This load torque signal is especially sensitive to faults, and a demonstration of the effect on and detection of compressor valve faults from the load torque signal is given. One of the key steps in the algorithm presented here is a procedure to "invert" the induction motor dynamic model equations to allow direct calculation of motor shaft speed and torque from stator current and voltage measurements. For this procedure a non-intrusive method to estimate motor model parameters from an in-situ induction motor driving a periodic load was developed.